A new enterprise storage lineup and client-side GPU-assisted acceleration aim to push AI workloads faster from dense cloud clusters to everyday laptops reshaping how organizations deploy and scale AI.
Phison’s latest storage and acceleration technologies arrive at a moment when enterprises are scrambling to match infrastructure with soaring AI compute demand. The company’s expanded portfolio anchored by two PCIe Gen5 enterprise SSDs and an iGPU-based AI acceleration layer targets AI training, inference, analytics and cloud-scale data operations with a focus on predictable performance and low latency.
The update centers on two SSD platforms built to handle high-volume, latency-sensitive AI pipelines. The higher-end model is engineered for throughput-heavy environments such as AI training clusters, HPC systems and large analytics engines. It delivers up to 14.5 GB/s reads, 12 GB/s writes and up to 3300K/1050K random IOPS, addressing the bottlenecks common in multi-node training and real-time analytics. Capacities reach 30.72 TB with endurance options up to 3 DWPD.
The key features are:
- Gen5 SSD throughput up to 14.5 GB/s read and 12 GB/s write
- Up to 3300K / 1050K random IOPS for AI-intensive workloads
- Capacities scaling to 30.72 TB (performance) and 15.36 TB (density)
- aiDAPTIV+ boosts iGPU AI agent performance by up to 25×
A second model targets hyperscale and cloud operators prioritizing density and energy-efficient performance for object storage, content delivery networks and large distributed databases. While using the same Gen5 backbone14.5 GB/s reads and 12 GB/s writes the design optimizes for compact E1.S deployments with capacities up to 15.36 TB.
Beyond the data center, the company is pushing AI acceleration deeper into the client stack. Its aiDAPTIV+ technology enables AI agents to run efficiently on integrated GPUs, delivering up to 25× faster performance and slashing inference response times dropping one test case from 73 seconds to 4 seconds. This intends to turn mainstream laptops into viable AI productivity machines for IT teams, developers and students without the cost of dedicated GPUs.
The rollout builds on earlier high-capacity Gen5 deployments, including a 122.88TB E3.L SSD now shipping to OEMs, with capacities up to 245TB slated for the portfolio. The goal: unify controller design, firmware and AI-oriented memory technologies under a single architecture that can scale from early AI experimentation to full-scale enterprise deployment.
